Abstarct: There are three approaches for word recognition: segmentation, holistic shape, and combination of them. At the first approach a word is broken into its letters, then each letter is recognized and the results combined together. In another scenario, the word image is given to the system for holistic recognition; i.e., recognizing the word as a whole shape. In many personal and official systems, document images are stored in low-resolution. This resolution is suitable for human reading, but it is hardly recognizable by machine. Even, most of Latin optical character recognition systems have been developed for 300 dots per inch. In low resolution documents, it is difficult to segment the word into its letters, so for this situation, holistic approach is adopted. In this thesis, recognition of the sub-words image in 96 dpi, baxsed on holistic shape is presented. In this presentation, the proposed system is baxsed on a three step method. In the first stage, the number of investigated sub-words in the dictionary is reduced by using clustering method; this strategy enhances the recognition speed as well as recognition performance. In the second stage, 4 nearest clusters of the test sub-word is found with a classifier and then by searching among that clusters, the 10 nearest sub-words to the test sample will be found. This process is repeated for all the sub-words of a word then at the last stage, the word will be recognized with probable occurrences sub-words sequence method. The accuracy which is estimated in this recognition algorithm is so convenient, it has the ability to recognize the words with more than one sub-word with recognition rate of 98.01% and recognize words in single sub-word with 82.53% rate.